Abstract: Financial futures market is an important part of modern financial market systems in China. However, financial derivatives have natural properties of high-yield and high risk.Once a futures risk event occurs, it will cause great destructive effect to the whole financial markets.So investors have always been paying great attention to the prevention of futures' overnight risk. However, little work has been done to detect volatility characteristics and risk features of overnight return.
By taking CSI300 stock index futures for sample in this paper,CAViaR model is adopted to directly calculate the common VaR of overnight return. Nevertheless, considering rare data available during optimization in extreme quantiles, the estimation results may be biased. Therefore, a new framework, which combining extreme value theory and CAViaR model, is built to estimate the extreme overnight risk and analyze the dynamic characteristic of different quantiles both in left tail and right tail.Then both Kupiec LR(likelihood ratio) test and dynamic quantile test are used to backtest the accuracy of these models.
The empirical results are summarized as follows: (1) overnight return exhibits stylized facts of positive skewness, leptokurtosis and non-normal distribution. But it lacks of long-term memory property. (2) The three CAViaR models have strong predictivity power to the common overnight risk, among which the AS model performs best, while there is no significant difference between SAV model and IGARCH model. (3) After adding the extreme theory to the CAViaR model, the newly-constructed CAViaR-EVT model still can accurately depict the dynamic process of overnight risk in extreme low quantiles. Moreover, its forecast results are more reasonable than EVT model and GARCH-EVT models.
Important practical and social implications are suggested. The CAViaR model and CAViaR-EVT model offer useful practical approaches to forecast futures' overnight risk. Moreover, it also provides a theoretical reference to carry out effective risk management and monitor activities for the Chinese stock index futures investors and regulators, such as position limits and margin ratio.

Key words: overnight risk, CAViaR model, CSI300 stock index futures, value at risk